Chainalysis finds 55% payment surge from China-linked wallets after US AI launches
A survey of AI middlemen helps explain the suspected connection. But payment timing alone cannot show who bought model access—or what they did with it.
Loading page…
A survey of AI middlemen helps explain the suspected connection. But payment timing alone cannot show who bought model access—or what they did with it.
Listen to this story
Chainalysis says payments from China-linked wallets to services selling routed access to US AI models rose 55% above baseline in the first three days after model launches. The pattern later reversed and recovered, but aggregate wallet activity cannot confirm what any customer bought or whether model responses were used for training. For investigators, the practical lead is to trace the brokers, access suppliers and payment processors behind the flows—not treat a launch-time spike as proof of individual activity.
In a survey of 100 AI proxy services, 40% accepted cryptocurrency, primarily stablecoins; the figure does not show what share of sales used crypto.
Payments fell to 21% below baseline around days 12–14, then reached 10% above baseline around days 18–20; Chainalysis linked the dip to Chinese open-source derivative release timing.
Resellers reportedly offer discounted credits for Claude, ChatGPT and Gemini, and some accept USDT on TRC20, BNB Smart Chain and Arbitrum.
US AI model launches leave a financial ripple in wallets that blockchain analytics firm Chainalysis identifies as China-linked. The firm found stablecoin payments running 55% above baseline during the first three days after launches. It associates that rise with middlemen selling restricted model access, while cautioning that the transactions cannot establish what individual customers purchased.
The services sit between customers and US model providers. Rather than buying a direct subscription, a customer can buy credits from an intermediary that routes requests through other accounts. BigGo Finance’s account describes services reselling access to Claude, ChatGPT and Gemini, typically through discounted credits.
Chainalysis describes a division of labor: access suppliers provide the connection to models, brokers connect buyers with suppliers, and payment processors handle the money. Some processors route stablecoin transfers through third parties to make them harder to trace. The customer-facing service and the payment destination can therefore involve different intermediaries.
Chainalysis surveyed 100 AI proxy services and found that 40% accepted cryptocurrency, primarily stablecoins. The finding measures payment acceptance among those services, not the share of their sales paid in crypto.
The payment options include USDT on three networks: TRC20, BNB Smart Chain and Arbitrum. Chainalysis also points to an economic incentive for buying access: it costs less than training comparable models from scratch. Customers can query a US model and use its responses to train a local system, a technique called distillation.
The timing is the distinctive finding. Chainalysis compared payment activity with US frontier-model launches and found that the early increases from China-linked wallets outpaced those across all wallets paying the services. The comparison concerns a post-launch change relative to baseline, rather than simply a large volume of crypto moving through the market.
The pattern did not remain a steady climb. Activity fell to 21% below baseline around days 12 to 14, then recovered to 10% above baseline during days 18 to 20. Chainalysis said the pullback coincided with the typical release window for Chinese open-source derivatives built from the queried models.
That interpretation has an important boundary: aggregate transactions cannot establish individual payment purposes. The timing can support Chainalysis’s association with proxy access, but it does not identify a particular purchase as model access or show that a buyer used responses for training. BigGo Finance also notes that the firm’s wallet-attribution methodology has not been publicly detailed.
An earlier Chainalysis study provides context for the size of the surrounding market. It counted $104.1 billion in self-custodied stablecoin transfers from China-linked wallets across 18.1 million transactions between July 2025 and June 2026. Those figures describe broader wallet activity; they are not an estimate of spending on AI proxies, model access or training.
For investigators, Chainalysis proposes following the connections among wallets, payment processors and trading platforms to identify infrastructure behind unauthorized model access. Its proposed next step is to map the intermediaries handling the money, rather than treat the post-launch payment spike alone as a complete account of the activity.
Loading discussion...
Join the conversation
Explain where a useful lead becomes too weak to act on.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.